In island systems with complex orography (e.g. Canary Islands), obtaining projections of climate extremes throughout the 21st century is necessary to evaluate the possible adverse effects of climate change. In this work, a dynamic downscaling methodology was applied to obtain the projections of temperature extremes indices. The WRF modeling system was properly configured with a spatial resolution of 3 km, for the periods: 2030–2059 (MID) and 2070–2099 (END), and for the RCPs 4.5 and 8.5 scenarios. This spatial-temporal resolution allows better modeling of the land-surface coupling processes (e.g., latent and sensible heat fluxes), which are one of the main sources of uncertainties in temperature extremes modeling. The initial and boundary conditions were defined by three CMIP5 Earth Systems Models: GFDL-ESM2M, MIROC-ESM, and IPSL-CM5. The future changes were calculated against the modeled reference period was 1980–2009 (HIS). The selected extremes indices were those defined by the Team of Experts on Climate Change Detection and Indices (ETCCDI) and were: monthly absolute maximum and minimum temperature respectively (TX and TN), monthly maximum of the diurnal temperature range (DTR), tropical nights (TR), warm days (TX90P), cold nights (TN10P), warm-spell duration index (WSDI) and cold-spell duration index (CSDI). Also, the return levels and return periods for annual maximum temperature were analyzed using the Generalized Extreme Value distribution (GEV).

The modeled indices were compared with those obtained from observations at nine ground-based stations for the HIS period. Despite the high spatial and temporal resolution of the models a bias is still observed between the modeled and observational values for the absolute indices, even when the simulations are driven by reanalysis data. However, the comparison of these indices around their previously unbiased means yields values on average of 0.85 with a standard deviation of 0.06 on a Perkins-based skill score. Regarding the 20-year return levels for maximum temperature, differences between the average of the models and observations are below 2 °C for all sites, except for the highest stations IZO and TFN, which reach 2.9 and 4.2 °C, respectively.

The analyses of the results indicate that the future projections of the indices obtained using any of the models remain constant from the mid-century to the end of the century for the RCP45 whereas they continue to increase if the RCP85 is considered. This finding shows that all models closely follow the variation in the CO₂-equivalent concentrations used as input. Thus, TX and TN are expected to increase, with an average change for the END period and RCP8.5 of 4.0 ± 0.5 °C and 4.4 ± 0.4 °C for TN. TX90p increases considerably (30 percentage points), and the TN10p index will be close to zero. The increase in temperatures is mainly due to, in addition to the modification of the synoptic patterns, a decrease in total cloud cover and soil moisture.

This decrease in soil moisture has a direct effect on the decrease in latent heat flux and an increase in sensible heat flux, associated with a projected increase of DTR. Also, the 20-year return levels for maximum temperature obtained for the HIS period will correspond to return periods between 1 and 6 years at the END period and RCP8.5.

在地形复杂的岛屿系统（例如加那利群岛）中，获得整个 21 世纪极端气候的预测对于评估气候变化可能产生的不利影响是必要的。在这项工作中，应用动态降尺度方法来获得极端温度指数的预测。WRF 建模系统在 2030-2059 (MID) 和 2070-2099 (END) 期间以及 RCP 4.5 和 8.5 情景中正确配置为 3 km 的空间分辨率。这种时空分辨率可以更好地模拟地表耦合过程（例如，潜热和感热通量），这是极端温度建模中不确定性的主要来源之一。初始条件和边界条件由三个 CMIP5 地球系统模型定义：GFDL-ESM2M、MIROC-ESM 和 IPSL-CM5。未来的变化是根据 1980-2009 年（HIS）的模拟参考期计算的。选定的极端指数是由气候变化检测和指数专家组 (ETCCDI) 定义的指数，分别是：月绝对最高和最低温度（TX 和 TN）、昼夜温度范围的月最大值（DTR）、热带夜晚(TR)、温暖的日子 (TX90P)、寒冷的夜晚 (TN10P)、暖期持续时间指数 (WSDI) 和寒期持续时间指数 (CSDI)。此外，使用广义极值分布 (GEV) 分析了年度最高温度的回归水平和回归期。月绝对最高和最低温度分别（TX 和 TN），昼夜温度范围的月最大值（DTR），热带夜晚（TR），温暖的日子（TX90P），寒冷的夜晚（TN10P），暖期持续时间指数（WSDI）和寒冷法术持续时间指数（CSDI）。此外，使用广义极值分布 (GEV) 分析了年度最高温度的回归水平和回归期。月绝对最高和最低温度分别（TX 和 TN），昼夜温度范围的月最大值（DTR），热带夜晚（TR），温暖的日子（TX90P），寒冷的夜晚（TN10P），暖期持续时间指数（WSDI）和寒冷法术持续时间指数（CSDI）。此外，使用广义极值分布 (GEV) 分析了年度最高温度的回归水平和回归期。

将模拟指数与 HIS 期间从九个地面站观测所得的指数进行了比较。尽管模型的空间和时间分辨率很高，但在绝对指数的建模值和观测值之间仍然存在偏差，即使模拟是由再分析数据驱动的。然而，将这些指数与之前的无偏平均值进行比较，得出的平均值为 0.85，基于 Perkins 的技能得分标准差为 0.06。关于最高温度的 20 年回归水平，除最高站 IZO 和 TFN 分别达到 2.9 和 4.2°C 外，所有站点的模型平均值与观测值之间的差异均低于 2°C。

对结果的分析表明，对于 RCP45，使用任何模型获得的指数的未来预测从本世纪中叶到本世纪末保持不变，而如果考虑 RCP85，它们会继续增加。这一发现表明，所有模型都密切关注用作输入的 CO _{2等效浓度的变化。}因此，TX 和 TN 预计会增加，END 期和 RCP8.5 的平均变化为 4.0 ± 0.5 °C 和 4.4 ± 0.4 °C 的 TN。TX90p 大幅增加（30 个百分点），TN10p 指数将接近于零。除了天气模式的改变外，温度的升高主要是由于总云量和土壤水分的减少。

土壤水分的这种减少对潜热通量的减少和感热通量的增加有直接影响，这与预计的 DTR 增加有关。此外，HIS 期间获得的最高温度的 20 年回归水平将对应于 END 期间和 RCP8.5 的 1 至 6 年之间的回归期。